Data Science Container

The data science container is a ready-to-use environment for machine learning, data analysis, and scientific computing. It includes a comprehensive Python stack, R, Julia, and GPU-accelerated libraries when provisioned with a GPU.

Pre-installed packages

Python data science: polars, pandas, numpy, scipy, scikit-learn, statsmodels, xgboost, lightgbm

Deep learning: PyTorch, transformers, tokenizers, accelerate

Visualization: matplotlib, seaborn, plotly, altair

GPU acceleration (when GPU attached): CUDA toolkit, RAPIDS (cuDF, cuML), PyTorch with CUDA support

Languages: Python 3.12, R, Julia

Tools: Jupyter (available but not the primary interface), pip, conda, git

When to use it

• Model training on large datasets with dedicated GPU resources
• Feature engineering and data wrangling with polars/pandas on high-memory instances
• Persistent development environment you can SSH into from your IDE
• Batch processing jobs that need significant compute
• GPU-accelerated workloads (RAPIDS for dataframe operations, PyTorch for training)

GPU support

Add a GPU at creation time for hardware-accelerated workloads:

curl -X POST https://console.carolinacloud.io/api/instance/ \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "resource_type": "container",
    "flavor": "datascience",
    "n_vcpus": 16,
    "mem_gib": 64,
    "disk_size_gib": 200,
    "gpu_model": "RTX 5090"
  }'

When a GPU is attached, CUDA, cuDNN, and GPU-accelerated Python libraries are available immediately. Verify with:

python3 -c "import torch; print(torch.cuda.is_available())"

Creating a data science container

# CPU-only
ccloud new container --cpus 8 --ram 32 --disk 100 --flavor datascience

# With GPU
ccloud new container --cpus 16 --ram 64 --disk 200 \
  --flavor datascience --name ml-training

Access

SSH only. No web interface. Connect your IDE via Remote-SSH for a full development experience.

ssh -p <port> ccloud@login.carolinacloud.io

If you want a browser-based notebook interface with the same data science stack, use the Data Science Notebook flavor instead.

AI integration

The Claude Code CLI is pre-installed. Pass an Anthropic API key at creation time to pre-authenticate it for the ccloud user. See AI Integration for details.

Auto-stop on idle

Training and analysis containers spend most of their time idle between jobs. Enable auto-stop to have the container shut itself off after a timeout of idle CPU — compute billing halts, data persists, and you can restart on demand.